English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

The state-of-the-art deep neural networks are vulnerable to commoncorruptions (e.g., input data degradations, distortions, and disturbancescaused by weather changes, system error, and processing). While much progresshas been made in analyzing and improving the robustness of models in imageunderstanding, the robustness in video understanding is largely unexplored. Inthis paper, we establish a corruption robustness benchmark, Mini Kinetics-C andMini SSV2-C, which considers temporal corruptions beyond spatial corruptions inimages. We make the first attempt to conduct an exhaustive study on thecorruption robustness of established CNN-based and Transformer-basedspatial-temporal models. The study provides some guidance on robust modeldesign and training: Transformer-based model performs better than CNN-basedmodels on corruption robustness; the generalization ability of spatial-temporalmodels implies robustness against temporal corruptions; model corruptionrobustness (especially robustness in the temporal domain) enhances withcomputational cost and model capacity, which may contradict the current trendof improving the computational efficiency of models. Moreover, we find therobustness intervention for image-related tasks (e.g., training models withnoise) may not work for spatial-temporal models.

Benchmarking the Robustness of Spatial-Temporal Models Against Corruptions